LITERATURE REVIEW OF RESPONSE RATES

advertisement
Appendix A
LITERATURE REVIEW OF RESPONSE RATES
In this appendix, we give a detailed account of the Internet survey
literature with respect to response rates in particular. The following
sections are structured around the number of response modes that
were used in each study (single response mode versus dual-response
mode) and by Web versus e-mail response modes.
Single Mode: Web
Studies on the use of the Web as a response mode vary widely in
terms of the nature of their target populations, how respondents are
recruited, and whether any attempts at statistical adjustment are
made in the studies’ analyses. In this section, we discuss surveys that
primarily or exclusively used the Web as the response mode. We
broadly classify these surveys by their method of recruiting respondents: through preselected samples, censuses and probability samples, convenience samples, or hybrid samples. By “preselected,” we
mean that the respondents were selected and screened to meet
specific criteria for responding to a Web survey. By “hybrid,” we refer
to various or multiple combinations of probability-based and
convenience-based methods used to recruit potential respondents.
For Web surveys that used either preselected or probability samples,
we compared outcomes in terms of the response rates that were
achieved. However, response rates cannot be computed when respondents are recruited through convenience sampling, such as
through various forms of advertising, or if the survey is simply posted
on the Web for anyone to complete. There are situations in which
convenience sampling does allow for the computation of completion
81
82
Conducting Research Surveys via E-Mail and the Web
rates, which is defined as the ratio of the number of surveys returned
to the number of requested surveys. However, in these cases, it is not
clear that a high completion rate conveys anything that is particularly meaningful or relevant about the respondents.
We begin by discussing one organization’s extensive efforts at evaluating the Web as a survey medium: the United States Census
Bureau’s Computer Assisted Survey Research Office (CASRO). The
U.S. Census Bureau has been actively engaged in research related to
electronic surveys (such as CSAQs) for the past decade. The Census
Bureau started using CSAQs in 1993 and 1994 by mailing DOS-based
diskettes to respondents. From 1996 to the present, CASRO has
fielded CSAQs by mailing Windows-based diskettes to respondents
and by sending CSAQs via the Web. The first Web CSAQ was conducted in 1997 and, as of this writing, nine more have been fielded,
four more are in production, and four are under development. Much
of CASRO’s completed work, which we concentrate on here, was directed toward business surveys, such as the Industrial Research and
Development survey.
Table A.1 lists Web surveys of preselected potential respondents,
which are from Sedivi Gaul (2001) and Nichols and Sedivi (1998).
With the exception of the Manufacturers’ Shipments, Inventories,
and Orders Survey, for which the survey team also sent a paper
survey in the initial mailing, these surveys were conducted solely via
the Web. The table shows that careful preselection can result in high
response rates. The exception is the 1998 Company Organization
Survey, for which the low response rate was attributed to the use of
an encryption level that resulted in many potential respondents not
being able to access the survey (168-bit encryption was used versus
the more common 128 bit). Lowering the encryption requirement to
128 bit in 1999 resulted in a significantly better response rate.
It is worth noting here that the U.S. Census Bureau put considerable
effort into carefully designing and extensively testing these instruments to make them as user friendly as possible. For example, the
Industrial Research and Development Survey instrument for the Web
was written completely in house in HTML and JavaScript. The program was designed perform real-time branching and editing,
opening in its own browser window with “help” information and edit
Literature Review of Response Rates
83
Table A.1
Response Rates for U.S. Census Bureau Web-Based Surveys
(Sent to Preselected Organizations)
Survey
Sample Size
Response Rate
Manufacturers’ Shipments, Inventories, and
73
89%
Orders Survey, 2000
Company Organization Survey, 1999
194
75%
Company Organization Survey, 1998
48
27%
a
Industrial Research and Development Survey,
50
68%
1997
aAfter the Web survey phase was completed, paper surveys were mailed to the nonrespondents. Thirteen respondents subsequently returned the paper version, for a
total response rate of 96 percent.
capabilities built in. Respondents could partially complete the survey, close the application, and then return later to continue the survey. The instrument had a menu bar on the right side of the screen
that permitted immediate branching to any section in the survey, so
respondents could choose to work through the instrument sequentially or jump around in any order they preferred. In addition to automatically writing the data to a database, the software also recorded
how respondents moved through the instrument—information that
could be used to improve future survey instruments. Nichols and
Sedivi (1998) provide a detailed description of the design and evaluation process.
Table A.2 presents results for other studies that used the Web as the
primary or only response mode and used censuses or probability
samples. This table shows more-modest response rates than those in
Table A.1.
Couper (2001) conducted an experiment in which 7,000 University of
Michigan students were randomized to receive a survey about drug
and alcohol use; 3,500 potential respondents received a mail survey
and 3,500 were notified of an equivalent Web-based survey.
Respondents in both groups received an incentive consisting of a $10
gift certificate. The Web-based survey achieved an almost 62-percent
response rate compared with a response rate of slightly less than 41
percent for the mail survey.
84
Conducting Research Surveys via E-Mail and the Web
Table A.2
Response Rates for Web-Based Surveys Using Censuses
or Probability Samples
Survey
Couper (2001)
Sample Size
7,000
a
Response Rate
62%
14,150
8%
Everingham (2001)
Jones and Pitt (1999)
b
Dillman et al. (1998)
1,298
200
9,522
44%
19%
41%
c
2,466
38%
Asch (2001)
Dillman et al. (1998)
Population
University of
Michigan students
College-bound high
school students and
college students
RAND employees
University staff
Purchasers of computer products
Purchasers of computer products
a
Most respondents were contacted via their parents, which reduced the response rate.
b
A relatively plain Web survey design was used in this experimental arm.
cA relatively elaborate Web survey design was used in this experimental arm.
Asch (2001) contacted a random sample of college-bound high
school youths and college students by mail (via their parents) and
asked them to participate in a Web survey. 1 Nonrespondents were
subsequently contacted in a follow-up mailing, which included a
mail survey. The study used incentives and several follow-ups,
including phone follow-ups to a subset of the sample. The overall
response rate was almost 21 percent, of which almost 8 percent
answered via the Web. This study is described in more detail in
Chapter Six.
Everingham (2001) conducted a “Work/Life Balance Survey” via the
Web in early 2000 at RAND. The survey consisted of slightly more
than 80 questions about quality-of-life programs. Respondents in
two geographically separate offices were initially contacted through
an e-mail that contained a link to the survey Web site. Ultimately, 44
percent of the eligible office staff members responded to the survey.
Because the target population was employees, Everingham was able
______________
1Contacting the sample was complicated and that complication impacted the overall
response rate for the survey. In general, the young adults were first contacted through
a letter sent to their parents at the parents’ last known home address. Parents were
then asked to forward the survey material to their sons and daughters.
Literature Review of Response Rates
85
to compare the demographics of respondents and nonrespondents
and the two groups were found to differ only in one dimension:
gender. A larger fraction of the respondents was female (59 percent)
compared with a fraction of the total population that was female (50
percent) and this difference is reasonably attributable to the survey
subject matter. While no equivalent paper-based survey was distributed to allow for direct response-rate comparisons, RAND had
previously fielded a paper-based survey on dependent care in 1990
that achieved a significantly higher response rate (more than 90 percent). Whether some or all of the differences in response rates between the two surveys can be attributed to differences in the survey
delivery mode cannot be determined.
Jones and Pitt (1999) sampled staff at ten universities whose staff directories were available on the Web. They compared three study
arms: contact and response by e-mail; contact by e-mail and response via the Web; and contact and response by mail. The response
rates for the three study arms were 34, 19, and 72 percent, respectively.
Dillman et al. (1998) conducted a survey of purchasers of computer
products who were at least 18 years of age and had used the Internet
from home, school, or work for at least one application other than email in the past month. Dillman et al. obtained a sample of purchasers of computer products and attempted to contact each potential respondent by phone up to five times. Those who agreed to
participate were then asked to respond on the Web and were sent an
incentive of two dollars. Respondents who initially agreed to participate but then did not were sent follow-up reminders by e-mail (their
e-mail addresses were obtained during the initial phone contact).
The study had two arms: one using a relatively plain Web survey design and one using a relatively fancy design. (The more-elaborate
surveys take longer to load on the computer.) The overall response
rates were 41 percent for the plain survey and 36 percent for the
fancy one. Dillman concluded that the plain design worked better
but also speculated that as Internet access speed increases, this difference may decrease significantly.
There are a number of studies that use convenience samples; often
respondents are recruited through advertisements of some form. As
we have noted in this report, for studies using convenience samples,
86
Conducting Research Surveys via E-Mail and the Web
response rates cannot be computed or are meaningless. Flemming
and Sonner (1999) reported on two Web surveys involving convenience samples. In one, individuals who visited the Pew Research
Center Web site were given an opportunity to fill out a Web survey. In
the other, respondents to an RDD phone survey were asked if they
were interested in participating in an unrelated Web survey. Because
both sets of respondents form convenience samples, response rates
are not given.
Kaye and Johnson (1999) conducted a Web survey about uses of the
Web for obtaining political information. Participants were recruited
through newsgroup postings and Web site links. In a social science
study of geographic mobility and other related topics, Witte et al.
(2000) recruited a large number of respondents: 32,688. The survey
design was unusual; it used a base module and four optional extension modules and respondents could elect to answer all of the extension modules.
Vehovar et al. (1999) conducted a large-scale survey targeted at the
Internet population of Slovenia, which corresponds to about 13 percent of the country’s total population. They sent out 19,000 e-mails
and advertised widely in traditional media. Coomber (1997) conducted a survey on the practices of illegal drug dealers. His target
population was dealers worldwide. Coomber solicited responses by
e-mail and through advertising, and collected responses on the Web
(with a very small number of respondents responding by mail) in the
hope that the participants would be encouraged to respond honestly
because of the perceived anonymity.
Dual Modes: Web and Mail
The number of studies that allow respondents to choose either a Web
or postal mail response mode is small. Nevertheless, these studies
are important because, for many populations, the fraction of respondents who can answer via the Web may not be sufficiently large
to make a Web response option economical, in which case mail is the
most-appropriate alternative mode. Table A.3 summarizes these
dual-mode studies, reporting the percentage of individuals who responded via the Web compared with the percentage that responded
by mail.
Literature Review of Response Rates
87
Table A.3
Studies Allowing Respondents to Choose a Web or Mail Response Mode
Total
Sample Size
9,596
Percentage Who
Chose to
Respond by …
Mail
Web
95%
5%
Study
Sedivi Gaul (2001)
and Griffin et
al. (2001)
(American
Community
Survey, 2000)
Sedivi Gaul (2001)
924
95%
5%
and Griffin et
al. (2001)
(Library Media
Center Survey,
1998)
Sedivi Gaul (2001)
13,440
81%
19%
and Griffin et
al. (2001)
(Library Media
Center Survey,
1999)
Quigley et al.
36,293
77%
23%
(2000) (DoD
study)
Quigley et al.
36,293
83%
27%
(2000) (DoD
study)
Zhang (2000)
201
20%
80%
a
Schleyer and
405
84%
16%
Forrest (2000)
aThe response mode in this case was either e-mail or fax.
Overall
Response
Rate
38%
Population
U.S. households
38%
Librarians
63%
Librarians
42%
37%
U.S. military and
spouses
Civilians
78%
74%
Researchers
Dentists
NOTE: The Quigley et al. entries represent two arms of the same study.
Table A.3 lists the results of two U.S. Census Bureau surveys from
Sedivi Gaul (2001) and Griffin et al. (2001)—the American Community Survey and the Library Media Center Survey. In contrast to
the respondents to the surveys listed in Table A.1, the respondents to
these surveys were not preselected and were also provided with a
paper survey. With these surveys, there is a definite negative effect on
Web response rates when respondents are not prescreened and
88
Conducting Research Surveys via E-Mail and the Web
when respondents are given another response mode as an alternative to the Web.
The results are mixed for the remaining studies listed in Table A.3. In
Zhang (2000) and Schleyer and Forrest (2000), respondents were initially contacted by e-mail; 80 percent of the respondents in the
Zhang study and 84 percent of the respondents in the Schleyer and
Forrest study responded via the Web. In contrast, the respondents in
the two arms of the Quigley et al. (2000) study were contacted via
mail; only 23 percent of the respondents in one arm of the study and
28 percent in the other responded via the Web. The studies of Zhang
and Schleyer and Forrest typically involve groups of respondents
who are largely or entirely computer literate and comfortable with
electronic communication. By comparison, the respondents in the
Quigley et al. study and American Community Survey study by Sedivi
Gaul and Griffin et al. tend to more closely approximate a general
cross-section of the U.S. public in terms of computer usage and familiarity.
Quigley et al. (2000) reported on a study by the U.S. Department of
Defense (DoD) involving a random sample of 36,293 participants, 52
percent of whom were active-duty personnel, 33 percent were military spouses, 9 percent were civilians working in military settings,
and 5 percent were reserve members. The study consisted of three
arms, two of which allowed respondents to reply by either mail or the
Web. In all cases, the respondents were initially contacted via postal
mail. The study design included a prenotification mailing and three
follow-up mailings.
In the Quigley et al. mail-with-Web-option study arm, paper surveys
were sent out (except with the prenotification) and each contact with
respondents (including the prenotification) provided the address for
responding via the Web if desired. The final response rate was 42
percent. Of those who responded, 23 percent chose to respond via
the Web and 77 percent by mail. In the Web-with-mail-option study
arm, respondents were expected to reply via the Web. They were also
given the option to request a paper survey, but very few people took
advantage of that option. Because of the poor response rate, a mail
survey was included with the third follow-up, which significantly
boosted the final response rate to 37 percent. Of those respondents,
27 percent chose to respond via the Web and 73 percent chose to
Literature Review of Response Rates
89
respond by mail, and most of that 73 percent responded only to the
third follow-up that included the paper survey. In contrast, the mailonly study arm had an overall response rate of 40 percent.
Zhang (2000) conducted a survey of researchers who were scheduled
to have their papers published in library science journals. The respondents, who were initially contacted via e-mail, could respond via
the Web or could request a mail survey. In the third of three followups, a paper survey was also sent by postal mail. The total sample
size was 201 and ultimately a 78 percent response rate was achieved.
Of that 78 percent, 80 percent chose to respond via the Web and 20
percent by mail.
The third follow-up generated more mail than e-mail responses, indicating that there is a slice in the target population who will not or
cannot fill out a Web survey and will not request a mail survey, but
will participate in a mail survey if the questionnaire is sent to them
directly. The total number of respondents to the third follow-up was
very small (roughly a dozen surveys were obtained by mail and a
half-dozen via the Web).
Not surprisingly, a comparison of respondents by response mode
showed that those who responded via the Web had a higher selfperceived overall ability to use the Internet, were using the Internet
more frequently, and were younger that those who responded by
mail. Nevertheless, some of the mail respondents were also highly
experienced Internet users.
Schleyer and Forrest (2000) assembled a convenience sample of 450
e-mail addresses in order to conduct a survey about clinical practices
among dentists. Schleyer and Forrest obtained the e-mail addresses
from large Internet discussion groups for dentists. Their survey consisted of 22 questions that were initially pilot tested. Nonrespondents
received three follow-up e-mail contacts. Schleyer and Forrest allowed respondents to return their surveys by the Web, e-mail, or fax;
they achieved an overall response rate of 74 percent.
Single Mode: E-mail
In general, the research comparing e-mail with other response
modes is limited, most likely because that mode was quickly eclipsed
90
Conducting Research Surveys via E-Mail and the Web
by the Web-based surveys. Most of the e-mail survey studies in the
literature are fairly limited in scope and weak in methodology and
can be characterized as simple pretests of a new medium rather than
careful experimental comparisons of a new survey mode versus existing modes.
However, the literature does contain some fairly rigorous attempts to
compare the response rates of surveys delivered via e-mail compared
with those delivered via traditional mail. As shown in Table A.4, surveys using e-mail as the sole response mode generally do not achieve
response rates equal to those of postal mail surveys.
Note that many of the studies listed in Table A.4 have relatively small
sample sizes and that as the e-mail response rate increases so does
the mail response rate. This suggests that the improved response
rates are likely attributable to an increased overall propensity of the
sample to respond because of differences in either the survey
methodology or the population. Only two of the studies (Couper et
al., 1999; Schaefer and Dillman, 1998) have relatively large sample
Table A.4
Studies Comparing Response Rates for E-Mail and Mail Response Modes
Total
Sample Size
400
500
418
Response Rate
E-Mail
Mail
Study Arm
Study Arm
6%
27%
7%
52%
19%
57%
Study
Population
Tse et al. (1995)
University staff
Tse (1998)
University staff
Schuldt and
MIS and marketing
Totten (1994)
faculty
Kittleson (1995)
153
28%
78%
Health educators
Jones and Pitt
200
34%
72%
University staff
(1999)
Mehta and
262
40%
45%
BBS newsgroup
Sivadas
users
(1995)
Couper et al.
8,000
43%
71%
Federal employees
(1999)
a
Schaefer and
904
58%
Washington State
53%
Dillman
University
(1998)
faculty
Parker (1992)
140
68%
38%
AT&T employees
aAnother 5 percent that were returned by mail are not included in this percentage.
Literature Review of Response Rates
91
sizes, and only one study (Parker, 1992) showed e-mail surveys resulting in a higher response rate than mail surveys. However, the
Parker study was conducted very early in the course of Internet surveying and its results are anomalous.
In a survey of administrative and teaching staff at the Chinese
University of Hong Kong, Tse et al. (1995) achieved only a 6-percent
response rate with an e-mail survey (with a sample size of 200) but a
27-percent response rate with a survey using university campus mail
(with a different sample of 200). In a follow-up experiment, Tse
(1998) randomly assigned 500 potential respondents selected from
the Chinese University telephone directory to receive either an email survey or an equivalent paper survey sent through the campus
mail. The result was a 7-percent response rate for the e-mail survey
and a 52-percent response rate for the mail survey. Tse et al. did find
an average initial response time of about one day for those who received an e-mail survey compared with an average response time of
2.5 days for those who received a paper survey through the campus
mail. The differences in the response times applied only to those who
responded to the first survey mailing and not the subsequent followup mailing.
The first Tse et al. mailing was followed by a second mailing to all 500
potential respondents, whether or not they had responded to the first
mailing. For the mail survey, 64 percent of those who responded via
campus mail did so after the first mailing, and the remaining 36 percent did so after the second mailing. In contrast, 86 percent of those
who responded via e-mail did so after the first mailing and only 14
percent responded after the second mailing. Thus, in this experiment, most e-mail survey recipients either responded almost immediately (within one day) or they did not respond at all.
Schuldt and Totten (1994), in surveying management information
system (MIS) and marketing faculty, achieved only a 19-percent response rate2 with e-mail (with a sample size of 218), as compared
with a 57-percent mail response rate (with a sample size of 200).
Similarly, in a comparison of e-mail versus postal mail surveys,
______________
2A total of 343 faculty members were in the initial e-mail sample. Of those, 125 were
undeliverable. If those 125 undeliverable e-mails are counted as nonrespondents, the
actual response rate is only 12 percent.
92
Conducting Research Surveys via E-Mail and the Web
Kittleson (1995) found that 153 health education professionals, each
receiving both a paper survey in the mail and a similar survey via email, were almost three times more likely to respond to the paper
survey as they were the e-mail survey (78 percent versus 28 percent).
Mehta and Sivadas (1995) also conducted an experiment involving email and postal mail surveys in which respondents were randomly
assigned to one of five groups: (1) those who were sent surveys via
regular mail with no prenotification and no reminders; (2) those who
were sent surveys via regular mail with prenotification and reminders; (3) those who were sent surveys via e-mail with no prenotification and no reminders; (4) those who were sent surveys via e-mail
with prenotification and reminders; and (5) an international group of
respondents who were sent surveys via e-mail with prenotification
and reminders. Group 2 also received a one-dollar incentive in the
survey package.
The most-direct comparison that can be made in the Mehta and
Sivadas experiment is between Group 1 (with a sample size of 202)
and Group 3 (with a sample size of 60), neither of which received
prenotifications or reminders. In those groups, mail surveys achieved
a 45-percent response rate and e-mail surveys achieved a 40-percent
response rate. A slightly less-direct comparison is between Group 2
and Groups 4 and 5; Group 2 achieved an 83-percent response rate
while Groups 4 and 5 achieved 63- and 64-percent response rates,
respectively. However, this comparison may not be a fair one
because Group 2 also received a one-dollar incentive whereas the email recipients did not.
A number of conclusions can be drawn from the Mehta and Sivadas
study. First, e-mail surveys do seem to benefit from advance notification and follow-up in the same way that mail surveys do; in this
case, these additional components increased the response rate almost 25 percentage points. Second, because researchers at this point
do not know how to most effectively employ incentives for surveys
that are conducted exclusively via the Internet, response rates for
Internet surveys may continue to lag until the effect of Internet survey incentives is better understood. In any case, employing incentives can prove to be very successful. In this experiment, the inclusion of just a one-dollar bill in the mail surveys increased response
rates by 20 percentage points.
Literature Review of Response Rates
93
In one of the few studies to randomize respondents to response
mode, Couper et al. (1999), in a survey of employees in federal statistical agencies, obtained an average response rate of about 43 percent
with e-mail compared with almost 71 percent with mail. The experiment conducted by Couper et al. randomized more than 8,000 employees of five different agencies. Couper et al. chose e-mail over the
Web as the survey mode because e-mail was almost universally
available in the five agencies whereas Web access was often not
available. The entire survey effort was carefully administered following TSD principles. In particular, advance notification was provided
for all surveys via agencywide e-mail broadcasts and bulletin board
notices, mail surveys were followed a week later by a postcard reminder, and e-mail surveys were followed a week later by an e-mail
reminder.
Schaefer and Dillman (1998), as reported in Dillman (2000), conducted an experiment involving e-mail versus postal mail surveys of
Washington State University faulty (with a survey sample numbering
904). Using a TDM approach, Schaefer and Dillman divided the potential respondents into four groups. The first group was contacted
by postal mail only (prenotification, survey, thank-you/reminder,
and replacement survey); the second group was contacted by e-mail
only; the third and fourth groups were contacted by a combination of
postal mail and e-mail. Schaefer and Dillman achieved a 58-percent
response rate with the all-postal-mail group. In comparison, they
achieved a 53-percent response rate with the all-e-mail group.
Most of the studies we examined conclude that mail achieves a
higher response rate than e-mail; Schaefer and Dillman (1998) and
Parker (1992) are the only studies we know of in which e-mail
achieved equal or higher response rates when compared with postal
mail. Parker conducted a survey of 140 former AT&T employees on
matters related to corporate policies for expatriation and repatriation. Parker reported a 63-percent response rate with e-mail (63 returned out of 100 sent by e-mail) compared with a 38-percent
response rate for postal mail (14 returned out of 40 sent by mail).
Interestingly, Parker attributed the difference in response rates to the
fact that, at the time, AT&T employees received a lot of corporate paper junk mail but little or no internal junk e-mail. Therefore, recipients of the paper survey were more likely to ignore the survey than
94
Conducting Research Surveys via E-Mail and the Web
were recipients of the e-mail version. With the spread of e-mail
spam, the situation is likely to be just the reverse today.
The only other published study that noted exceptional response rates
with e-mail is Walsh et al. (1992) in which potential respondents
were solicited by e-mail and were offered the option to respond by email or request a paper survey by postal mail. Although Walsh et al.
did not conduct an equivalent postal-mail-only survey for comparison, for an e-mail survey of a random sample of scientific computer
network subscribers (300 from a total population of 1,100), they
achieved a 76-percent overall response rate. Walsh et al. also sent
nonrespondents two follow-up reminders and employed a lottery
prize of $250 as an incentive.
Walsh et al. found that 58 percent of the random sample replied by email and 18 percent responded by postal mail. They also received requests from an additional 104 subscribers (who were not chosen in
the sample of 300) to participate in the survey. Of the self-selected
104 subscribers, 96 percent responded by e-mail. Not surprisingly,
Walsh et al. also found a positive correlation between a respondent’s
propensity to respond electronically and the amount of the respondent’s network usage.
Multiple Modes: Web or E-Mail and Telephone
We found no studies that evaluate mixed modes using either the Web
and the telephone or e-mail and the telephone. This is not particularly surprising given that Web surveys are often used to reduce survey costs and interviewing by telephone is very expensive. However,
telephone contact or response may have other benefits, such as improving response rates, and deserves study in this regard.
Download
Related flashcards
Web colors

17 Cards

Web colors

17 Cards

Web design

12 Cards

Create flashcards